Difference between revisions of "Xen 4.3 NUMA Aware Scheduling"

From Xen
(Updated the content of the early perf. benchmarks section)
Line 8: Line 8:
 
This document aims at describing what was included, regarding NUMA aware scheduling, in <em>Xen 4.3</em>. You can find other articles about NUMA in the [[:Category:NUMA|NUMA category]].
 
This document aims at describing what was included, regarding NUMA aware scheduling, in <em>Xen 4.3</em>. You can find other articles about NUMA in the [[:Category:NUMA|NUMA category]].
   
  +
= Preliminary/Exploratory Work =
= NUMA Aware Scheduling: Description and Performance Impact =
 
   
 
Suppose we have a VM with all its memory allocated on NODE#0 and NODE#2 of our NUMA host. One may think that the best thing to do would be to pin the VM’s vCPUs on the pCPUs related to the two nodes. However, pinning is quite unflexible: what if those pCPUs get very busy while there are completely idle pCPUs on other nodes? It will depend on the workload, but it is not hard to imagine that having some chance to run --even if on a remote node-- would be better than not running at all.
 
Suppose we have a VM with all its memory allocated on NODE#0 and NODE#2 of our NUMA host. One may think that the best thing to do would be to pin the VM’s vCPUs on the pCPUs related to the two nodes. However, pinning is quite unflexible: what if those pCPUs get very busy while there are completely idle pCPUs on other nodes? It will depend on the workload, but it is not hard to imagine that having some chance to run --even if on a remote node-- would be better than not running at all.
Line 24: Line 24:
 
The full set of results for these early benchmarks is available [http://xenbits.xen.org/people/dariof/benchmarks/specjbb2005-numa/ here]. There was a blog post about this, and it is still online at [http://blog.xen.org/index.php/2012/05/16/numa-and-xen-part-ii-scheduling-and-placement/ this address].
 
The full set of results for these early benchmarks is available [http://xenbits.xen.org/people/dariof/benchmarks/specjbb2005-numa/ here]. There was a blog post about this, and it is still online at [http://blog.xen.org/index.php/2012/05/16/numa-and-xen-part-ii-scheduling-and-placement/ this address].
   
  +
= The Actual Solution in Xen 4.3 =
== NUMA Placement and Scheduling ==
 
So, here's the situation: automatic initial placement has been [[Xen_4.2_Feature_List|included in Xen 4.2]], inside libxl. This means, when a VM is created (of course, if that happens through libxl) a set of heuristics decide on which NUMA node his memory has to be allocated, and the vCPUs of the VM are <strong>statically pinned</strong> to the pCPUs of such node.
 
   
On the other hand, NUMA aware scheduling has included in Xen 4.3. This mean, instead of being statically pinned, the vCPUs of the VM will <strong>strongly prefer</strong> to run on the pCPUs of the NUMA node, but they can run somewhere else as well... And this is what this status report is all about.
+
Automatic placement made it to [[Xen_4.2_Feature_List|Xen 4.2]], and that meant, when a VM is created, a (set of) NUMA node(s) is picked to store its memory, and its vCPUs <strong>statically pinned</strong> to the pCPUs of such node(s). With NUMA aware scheduling, which was included in [[Xen_4.3_Feature_List|Xen 4.3]], the latter is no longer the case. In fact, instead of using pinning, the vCPUs <strong>strongly prefers</strong> to run on the pCPUs of the NUMA node(s), but they can run somewhere else as well.
   
  +
During development, more benchmarks were run. For example the following ones:
== Performance Numbers? ==
 
  +
* SpecJBB: this is all about throughput, thus pinning is likely the ideal solution;
Sure thing! Benchmarks similar to the ones already described in the past have been performed. More specifically, directly from the cover letter of the Xen 4.3 implementation. Here's what has been done:
 
  +
* Sysbench-memory: this is the time it takes for writing a fixed amount of memory (and then it is the throughput that is measured). What we expect is locality to be important, but at the same time the potential imbalances due to pinning could have a say in it;
<blockquote>
 
  +
* LMBench-proc: this is the time it takes for a process to fork a fixed number of children. This is much more about latency than throughput, with locality of memory accesses playing a smaller role and, again, imbalances due to pinning being a potential issue.
<pre>I ran the following benchmarks (again):
 
  +
The host was a 2 NUMA box, where 2 to 10 VMs (2 vCPUs and 960 RAM each) were executing the various benchmarks concurrently. The results looks as follows:
* SpecJBB is all about throughput, so pinning is likely the ideal
 
  +
<pre>
solution.
 
  +
----------------------------------------------------
* Sysbench-memory is the time it takes for writing a fixed amount
 
of memory (and then it is the throughput that is measured). What
 
we expect is locality to be important, but at the same time the
 
potential imbalances due to pinning could have a say in it.
 
* LMBench-proc is the time it takes for a process to fork a fixed
 
number of children. This is much more about latency than
 
throughput, with locality of memory accesses playing a smaller
 
role and, again, imbalances due to pinning being a potential
 
issue.</pre>
 
</blockquote>
 
This all happened on a 2 node host, where 2 to 10 VMs (2 vCPUs and 960 RAM each) were executing the various benchmarks concurrently. Here they are the results:
 
<blockquote>
 
<pre> ----------------------------------------------------
 
 
| SpecJBB2005, throughput (the higher the better) |
 
| SpecJBB2005, throughput (the higher the better) |
 
----------------------------------------------------
 
----------------------------------------------------
Line 69: Line 56:
 
| 6 | 986.44955 | 1076.7447 | 900.21504 |
 
| 6 | 986.44955 | 1076.7447 | 900.21504 |
 
| 10 | 1211.2434 | 1371.6014 | 1285.5947 |
 
| 10 | 1211.2434 | 1371.6014 | 1285.5947 |
----------------------------------------------------</pre>
+
----------------------------------------------------
</blockquote>
+
</pre>
  +
Which, reasoning in terms of %-performance increase/decrease, means NUMA aware
 
scheduling does as follows, as compared to no-affinity at all and to static pinning:
+
Which, reasoning in terms of %-performance increase/decrease, means NUMA aware scheduling does as follows, as compared to no-affinity at all and to static pinning:
  +
<blockquote>
 
  +
<pre>
<pre> ----------------------------------
 
  +
----------------------------------
| SpecJBB2005 (throughput) |
 
  +
| SpecJBB2005 (throughput) |
----------------------------------
 
  +
----------------------------------
| #VMs | No affinity | Pinning |
 
| 2 | +13.05% | +0.21% |
+
| #VMs | No affinity | Pinning |
| 6 | +12.30% | +0.53% |
+
| 2 | +13.05% | +0.21% |
| 10 | +4.31% | -8.82% |
+
| 6 | +12.30% | +0.53% |
  +
| 10 | +4.31% | -8.82% |
----------------------------------
 
  +
----------------------------------
| Sysbench memory (throughput) |
 
  +
| Sysbench memory (throughput) |
----------------------------------
 
  +
----------------------------------
| #VMs | No affinity | Pinning |
 
| 2 | +15.44% | +3.79% |
+
| #VMs | No affinity | Pinning |
| 6 | +11.24% | +5.72% |
+
| 2 | +15.44% | +3.79% |
| 10 | +4.18% | -1.34% |
+
| 6 | +11.24% | +5.72% |
  +
| 10 | +4.18% | -1.34% |
----------------------------------
 
  +
----------------------------------
| LMBench proc (latency)  |
 
  +
| LMBench proc (latency)  |
| NOTICE: -x.xx% = GOOD here |
 
  +
| NOTICE: -x.xx% = GOOD here |
----------------------------------
 
  +
----------------------------------
| #VMs | No affinity | Pinning |
 
  +
| #VMs | No affinity | Pinning |
----------------------------------
 
  +
----------------------------------
| 2 | -5.66% | -0.50% |
 
| 6 | -9.58% | -19.61% |
+
| 2 | -5.66% | -0.50% |
| 10 | +5.78% | -6.69% |
+
| 6 | -9.58% | -19.61% |
  +
| 10 | +5.78% | -6.69% |
----------------------------------</pre>
 
  +
----------------------------------
</blockquote>
 
  +
</pre>
The tables show how, when not in overload (where overload='more vCPUs than pCPUs'), NUMA scheduling is <span style="color: #000000;"><strong>the absolute best</strong></span>. In fact, not only it does a lot better than no-pinning on throughput biased benchmarks, as well as a lot better than pinning on latency biased benchmarks (especially with 6 VMs), it also equals or beats both under adverse circumstances (adverse to NUMA scheduling, i.e., beats/equals pinning in throughput benchmarks, and beats/equals no-affinity on the latency benchmark).
 
  +
  +
The tables show how, when not in overload (where overload='more vCPUs than pCPUs'), NUMA aware scheduling is <span style="color: #000000;"><strong>the absolute best</strong></span>. In fact, not only it does a lot better than no-pinning on throughput biased benchmarks, and a lot better than pinning on latency biased benchmarks (especially with 6 VMs), it also equals or beats both under adverse circumstances (adverse to NUMA aware scheduling, i.e., beats/equals pinning in throughput benchmarks, and beats/equals no-affinity on the latency benchmark).
   
When the system is overloaded, NUMA scheduling scores in the middle, as it could have been expected. It must also be noticed that, when it brings benefits, they are not as huge as in the non-overloaded case. However, this only means that there is still room for more optimization, right?  In some more details, the current way a pCPU is selected for a vCPU that is waking-up, couples particularly bad with the new concept of NUMA node affinity. Changing this is not trivial, because it involves rearranging some locks inside the scheduler code, but is already being worked-on.
+
When the system is overloaded, NUMA scheduling scores in the middle, as it could have been expected. It must also be noticed that, when it brings benefits, they are not as huge as in the non-overloaded case (which probably means there is still room for some more optimization). In particular, the current way a pCPU is selected, when a vCPU waks-up, couples particularly bad with the new concept of NUMA affinity. Changing this is not trivial, because it involves rearranging some locks inside, but can be done, if deemed worthwhile.
Anyway, even with what we have right now, we are <em>overloading the test box by 20%</em> here (without counting Dom0 vCPUs!) <em>and still seeing improvements</em>, which is definitely not bad!
 
   
The [http://xenbits.xen.org/docs/unstable/misc/xl-numa-placement.html in tree documentation] has some more details about NUMA aware scheduling, and the interactions it has with [[Xen_4.2_Automatic_NUMA_Placement|automatic NUMA placement]].
+
The [http://xenbits.xen.org/docs/unstable/misc/xl-numa-placement.html in tree documentation] has some more details about NUMA aware scheduling, and the interactions it has with [[Xen_4.2_Automatic_NUMA_Placement|automatic NUMA placement]]. Also, there was a blog post about this topic too, available [http://blog.xen.org/index.php/2013/03/14/numa-aware-scheduling-development-report/ here].
   
 
[[Category:Xen]]
 
[[Category:Xen]]

Revision as of 18:21, 21 February 2014

When dealing with NUMA machines, it is (among other things) very important that we:

  1. achieve a good initial placement, when creating a new VM;
  2. have a solution that is both flexible and effective enough to take advantage of that placement during the whole VM lifetime.

The latter, which basically, means: <<When starting a new Virtual Machine, to which NUMA node should I "associate" it with?>>. The latter is more about: <<How hard should the VM be associated to that NUMA node? Could it, perhaps temporarily, run elsewhere?>>, is what is usually called NUMA aware scheduling.

This document aims at describing what was included, regarding NUMA aware scheduling, in Xen 4.3. You can find other articles about NUMA in the NUMA category.

Preliminary/Exploratory Work

Suppose we have a VM with all its memory allocated on NODE#0 and NODE#2 of our NUMA host. One may think that the best thing to do would be to pin the VM’s vCPUs on the pCPUs related to the two nodes. However, pinning is quite unflexible: what if those pCPUs get very busy while there are completely idle pCPUs on other nodes? It will depend on the workload, but it is not hard to imagine that having some chance to run --even if on a remote node-- would be better than not running at all.

The idea is, then, to give the scheduler some hints about where a VM’s vCPUs should be executed (and this preference, in this context, will be called from now on NUMA affinity). It then can try at its best to honor these suggestions of ours, but not at the cost of subverting its own algorithm. Here they are some early experimental results for this idea (dating back to [http://lists.xen.org/archives/html/xen-devel/2012-04/msg00732.html this patchset). The various curves in the graph below represents the throughput achieved one VM when it is:

  • scheduled without any pinning or NUMA affinity, i.e., cpus="all" in the config file (the red line);
  • pinned on NODE#0, so that all its memory accesses are local (the green line);
  • scheduled with NUMA affinity set to NODE#0, and no pinning, (blue line).

The plot shows is the percent increase of each configuration with respect to the worst possible case (i.e., when all memory access are remote).

kernbench_avg2.png

It appears quite clear that, introducing NUMA affinity increases performance by ~12% to ~18% from the worst case. It enables up to ~8% performance increase, as compared to unpinned behavior, and that the higher the load on the host, the better.

The full set of results for these early benchmarks is available here. There was a blog post about this, and it is still online at this address.

The Actual Solution in Xen 4.3

Automatic placement made it to Xen 4.2, and that meant, when a VM is created, a (set of) NUMA node(s) is picked to store its memory, and its vCPUs statically pinned to the pCPUs of such node(s). With NUMA aware scheduling, which was included in Xen 4.3, the latter is no longer the case. In fact, instead of using pinning, the vCPUs strongly prefers to run on the pCPUs of the NUMA node(s), but they can run somewhere else as well.

During development, more benchmarks were run. For example the following ones:

  • SpecJBB: this is all about throughput, thus pinning is likely the ideal solution;
  • Sysbench-memory: this is the time it takes for writing a fixed amount of memory (and then it is the throughput that is measured). What we expect is locality to be important, but at the same time the potential imbalances due to pinning could have a say in it;
  • LMBench-proc: this is the time it takes for a process to fork a fixed number of children. This is much more about latency than throughput, with locality of memory accesses playing a smaller role and, again, imbalances due to pinning being a potential issue.

The host was a 2 NUMA box, where 2 to 10 VMs (2 vCPUs and 960 RAM each) were executing the various benchmarks concurrently. The results looks as follows:

 ----------------------------------------------------
 | SpecJBB2005, throughput (the higher the better)  |
 ----------------------------------------------------
 | #VMs | No affinity |  Pinning  | NUMA scheduling |
 |    2 |  43318.613  | 49715.158 |    49822.545    |
 |    6 |  29587.838  | 33560.944 |    33739.412    |
 |   10 |  19223.962  | 21860.794 |    20089.602    |
 ----------------------------------------------------
 | Sysbench memory, throughput (the higher the better)
 ----------------------------------------------------
 | #VMs | No affinity |  Pinning  | NUMA scheduling |
 |    2 |  469.37667  | 534.03167 |    555.09500    |
 |    6 |  411.45056  | 437.02333 |    463.53389    |
 |   10 |  292.79400  | 309.63800 |    305.55167    |
 ----------------------------------------------------
 | LMBench proc, latency (the lower the better)     |
 ----------------------------------------------------
 | #VMs | No affinity |  Pinning  | NUMA scheduling |
 ----------------------------------------------------
 |    2 |  788.06613  | 753.78508 |    750.07010    |
 |    6 |  986.44955  | 1076.7447 |    900.21504    |
 |   10 |  1211.2434  | 1371.6014 |    1285.5947    |
 ----------------------------------------------------

Which, reasoning in terms of %-performance increase/decrease, means NUMA aware scheduling does as follows, as compared to no-affinity at all and to static pinning:

 ----------------------------------
 | SpecJBB2005 (throughput)       |
 ----------------------------------
 | #VMs | No affinity |  Pinning  |
 |    2 |   +13.05%   |  +0.21%   |
 |    6 |   +12.30%   |  +0.53%   |
 |   10 |    +4.31%   |  -8.82%   |
 ----------------------------------
 | Sysbench memory (throughput)   |
 ----------------------------------
 | #VMs | No affinity |  Pinning  |
 |    2 |   +15.44%   |  +3.79%   |
 |    6 |   +11.24%   |  +5.72%   |
 |   10 |    +4.18%   |  -1.34%   |
 ----------------------------------
 | LMBench proc (latency)         |
 | NOTICE: -x.xx% = GOOD here     |
 ----------------------------------
 | #VMs | No affinity |  Pinning  |
 ----------------------------------
 |    2 |    -5.66%   |  -0.50%   |
 |    6 |    -9.58%   | -19.61%   |
 |   10 |    +5.78%   |  -6.69%   |
 ----------------------------------

The tables show how, when not in overload (where overload='more vCPUs than pCPUs'), NUMA aware scheduling is the absolute best. In fact, not only it does a lot better than no-pinning on throughput biased benchmarks, and a lot better than pinning on latency biased benchmarks (especially with 6 VMs), it also equals or beats both under adverse circumstances (adverse to NUMA aware scheduling, i.e., beats/equals pinning in throughput benchmarks, and beats/equals no-affinity on the latency benchmark).

When the system is overloaded, NUMA scheduling scores in the middle, as it could have been expected. It must also be noticed that, when it brings benefits, they are not as huge as in the non-overloaded case (which probably means there is still room for some more optimization). In particular, the current way a pCPU is selected, when a vCPU waks-up, couples particularly bad with the new concept of NUMA affinity. Changing this is not trivial, because it involves rearranging some locks inside, but can be done, if deemed worthwhile.

The in tree documentation has some more details about NUMA aware scheduling, and the interactions it has with automatic NUMA placement. Also, there was a blog post about this topic too, available here.